UCT/IB/ZE: Enable GPUDirect RDMA for Intel Xe devices#11187
Merged
Conversation
Add Intel Xe driver detection in IB MD for GDR capability. Extend ZE copy MD access_mem_types to include HOST memory to support GDR memory registration paths. Signed-off-by: Yihua Xu <yihua.xu@intel.com> Signed-off-by: Yaser Afshar <yaser.afshar@intel.com>
tvegas1
approved these changes
Feb 17, 2026
yosefe
approved these changes
Feb 17, 2026
This was referenced Feb 25, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What?
Revives #11005 with minor cleanup: Add GPUDirect RDMA (GDR) support for Intel Xe devices by:
/sys/module/xe/srcversion) in IB MD to enable GDR capability forUCS_MEMORY_TYPE_ZE_DEVICEaccess_mem_typesto includeUCS_MEMORY_TYPE_HOSTto allow IB transports to access host memory through ZE MD registrationWhy?
In GDR mode, eliminating extra memcpy operations between host staging buffers and GPU memory improves bandwidth and reduces latency for RDMA transfers on Intel GPU hardware. Currently, the IB layer does not recognize Intel Xe devices for GPUDirect, and the ZE copy MD does not expose HOST memory access required for GDR interoperability.
How?
src/uct/ib/base/ib_md.c): Adduct_ib_check_gpudirect_driver()call to probe/sys/module/xe/srcversionalongside existing CUDA and ROCm checks. This sets the appropriate bit inmd->reg_mem_typeswhen the Xe kernel module is loaded.src/uct/ze/copy/ze_copy_md.c): AddUCS_BIT(UCS_MEMORY_TYPE_HOST)tomd_attr->access_mem_typesalongside existing ZE memory types. This allows IB transports to register host memory through the ZE MD for GDR operations.Related: #11180 (UCT/ZE: Add device topology registration) - Complementary ZE infrastructure changes. These PRs are independent but both advance ZE/IB integration.
Signed-off-by: Yihua Xu yihua.xu@intel.com
Signed-off-by: Yaser Afshar yaser.afshar@intel.com